Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!world!decwrl!sgi!vjs@rhyolite.wpd.sgi.com
From: vjs@rhyolite.wpd.sgi.com (Vernon Schryver)
Newsgroups: comp.dcom.lans
Subject: Re: Ethernet chip performance request
Message-ID: <83987@sgi.sgi.com>
Date: 4 Feb 91 23:28:04 GMT
References: <4467@alliant.Alliant.COM>
Sender: guest@sgi.sgi.com
Organization: Silicon Graphics, Inc., Mountain View, CA
Lines: 34

In article <4467@alliant.Alliant.COM>, werme@Alliant.COM (Ric Werme) writes:
> I recently got a BSD driver for Rockwell/CMC's CMC-130 Ethernet board running
> at 1.1 MB/sec for UDP data and 0.8 MB/sec for TCP.  (If I get a chance, I'll
> work on the latter some.)  So far we haven't run into any problems talking to
> systems that can't handle the minimal delay between packets.  However,
> Van Jacobson's reports of problems with Intel 82586 performance leaves me
> worried about other systems our customers may hookup.  Also, I think it was
> here that some people described a PC system that needed substantial delays
> between IP fragments.
>  ...


Delays between IP packets are commonly not needed for TCP, provided the TCP
MTU is reasonable and not causing IP fragmentation, and provided the
transmitter is reasonable.

The familiar NFS drill with slow PC's is to make the buffer size small
enough to prevent UDP/IP fragmentation.  There are always many systems at
Connectathon that require this.

There are many standard, relatively low cost workstations that cause ttcp
to report > 1Mbyte/sec over ethernet.  A concrete example is an IRIS 4D25
running IRIX 3.3.2.  There are doubtless many others.  I mention the IRIS
only so I can note that we've been shipping >=800KB for a long time, and
the only troubles reported by customers concerning slow systems have been
fixed by adjusting rsize and wsize as noted above.

Please note that 800KB for a third party board is a respectible number.
The FXP is a respectible board.  The person above and CMC are to be
congratulated.



Vernon Schryver,